The Met office CO2 concentration forecast
Duncan Golicher
4/21/2023
Introduction
The UK meteorological office have published a rather odd document in which they forecast the global atmospheric carbon dioxide content in 2023 which they link to the El Nino Southern Oscillation (ENSO).
https://www.metoffice.gov.uk/research/climate/seasonal-to-decadal/long-range/forecasts/co2-forecast
Some background to my interest in the effects of ENSO. In 1998 a particularly intense El Nino event occurred. At the same time, Southern Mexico experienced a delayed start to the rainy season. I was working on forests impacted by drought, so I was interested in the effect after finishing a PhD. I wrote a paper which looked closely at the historical correlation between the two events (Golicher, Ramirez-Marcial, and Levy Tacher 2006). Although there was some evidence of a link, it was not strong enough to be used for prediction. This was stated as the conclusion to the work.
It is unclear why an institution primarily involved in local weather forecasting would take on this extremely complex research task. The year on year increase in atmospheric Co2 is one of the most consistent trends known. There would therefore seem to be little need to forecast it. On the other hand, the link between carbon emissions and the ENSO is still uncertain.
The introduction to the report is intriguing.
“Carbon dioxide will continue to build up in the atmosphere in 2023 due to ongoing emissions from fossil fuel burning, land use change and cement production. However, this year’s annual CO2 rise is once again predicted to be smaller than the average for 2011-2020 for the third year running, due to an extended but temporary increase in the removal of carbon from the atmosphere by tropical vegetation in response to weather conditions associated with the third consecutive year with La Niña conditions. Without the temporary effect of La Niña, the 2023 CO2 rise would have been faster than that required to track a scenario to limit global warming to 1.5°C.”
So the main purpose of the research appears to link changes in the rate of CO2 accumulation in the atmosphere to the El Nino Southern Oscillation. There is an alternative explanation. This would be that emissions fell as a result first of pandemic responses and subsequently high energy prices. Yet this was not considered, possibly because direct emissions can be measured independently and have not shown a major fall.
Linking changes in C02 to the ENSO is a very difficult task. The chain of causality is extremely long and very tenuous. First the changes in sea surface temperature have to effect terrestrial climate. This does occur locally, but the global extent of the effect is important in this case. Then any change in climate must effect the growth of vegetation. Finally this must result in a change in the carbon balance. Each step is not easy to predict.
The report is written in a very strident journalistic style and makes assertions without documenting all its data sources and assumptions.
“We predict this seasonal cycle to peak at a monthly mean value of 423.3 ± 0.5 ppm in May. From comparison with reconstructions of past CO2 levels from isotopes of carbon and boron in marine sediments, this will be the highest atmospheric CO2 concentration for over 2 million years.”
This comment seemed particularly odd.
“Despite being a La Niña year with a relatively small CO2 increase compared to recent years, the central estimate of the forecast CO2 rise of 1.97 ± 0.52 ppm for 2022-2023 would be the 21st largest annual increment in the Mauna Loa record of 64 years”
I.e. the predicted rise, which has not yet happened, will be unusually high even though the recent measured rises were unusually low.
Annual fluctuations in atmospheric CO2 concentrations occur as vegetation in the northern hemisphere absorbs carbon dioxide during the growing season which is released during the winter when vegetation decomposes. This effect does not take place in the aseasonal equatorial forests, although at higher latitudes around the tropics prolonged dry seasons do play a role in the carbon cycle.
However, the met office report claims
“This suggests that without a La Niña response in the atmosphere and tropical land ecosystems, the forecast annual mean CO2 rise from 2022 to 2023 would be 2.32 ppm, a faster rise than the 1.97 ppm forecast when accounting for La Niña.” This is a very bold claim to make.
The whole issue is clearly highly nuanced and uncertain. Making such bold assertions stands in contrast to previous careful scientific work that seeks to understand the complexity involved in the annual and interannual fluctuations in atmospheric CO2 (Bacastow et al. 1980). A large proportion of the seasonally fluctuating uptake of atmospheric carbon dioxide by terrestrial vegetation seems to occur in boreal forests, but much less in tropical forests (Tagesson et al. 2020) . Thus unless the ENSO effect extends to the Taiga it would apparently be unable to play more than a very minor role.
The data are therfore worth looking at independently.
Looking at the data
The report’s authors obtained data from the Scripps institute. It can be downloaded from here.
https://scrippsco2.ucsd.edu/data/atmospheric_co2/
The daily flask readings from Mauna Loa were read into R.
d<- read_csv("daily_flask_co2_mlo.csv",
col_names = FALSE, skip = 69)[,c(1,7)]
names(d)<-c("date","co2")
d$date<-as.Date(d$date)
d$year<-year(d$date)
d$week<-week(d$date)A simple statistical model can be fit to the data using a general additive model which takes the week of the year and the year as input.
d$trend<-predict(gam(data=d,co2~s(week)+s(year)))Notice that there are always a few spuriously high values in the raw daily flask data. This is particularly noticeable if the earliest part of the time series is looked at (double click the dygraph to zoom out). This has always been something of a challenge for those ensuring the quality of the data, as the errors are asymmetrical. It is hard to get spuriously low readings. Higher than expected readings are likely to be due to some local source of carbon dioxide, such as volcanic eruptions in Hawaii. I’ll assume that quality control has already screened out most of these.
This purely empirical model does fit the data very well. It can be easily extended outside the range of the data in order to backcast and forecast. Neither of these procedures are recommended practice. An empirical data fit should be bound within the range of the data used to fit it. Such a model should not be used for prediction. However it can be tried to see what the results are.
d<-filter(mlo,date<dmy("01-01-2023"))
mod<-gam(data=d,co2~s(week)+s(year,k=5))
d<-data.frame(date=seq(as.Date("1900/1/1"), as.Date("2100/1/1"), "weeks"))
d$year<-year(d$date)
d$week<-week(d$date)
d$trend<-predict(mod,newdata=d)
dygraph(xts(data.frame(trend=d$trend),d$date), main = "Modelled Co2 measurements from Mauna Loa", xlab = "Date", ylab = "CO2 in parts per million") %>% dyRangeSelector(dateWindow = c("1900-01-01", "2100-01-01")) The backcast using the empirical model fit using data from 1960 to 2022 predicts a preindustrial CO2 concentration that more or less matches the conventionally agreed figure of around 290 parts per million. If the emiprical model prediction is taken as a “business as usual scenario” the CO2 concentration would be doubled before the end of the century.
Zooming in on the real time series shows an element that received only cursory attention in the Met office report.
This is very surprising. Despite the worldwide reduction in travel and some non essential consumption in 2020 there was no notable deviation from the previous trend. As reducing emissions through adopting “net zero” policies has become a hot issue, this counterintuitive observation may have motivated the Met office to come up with an alternative explanation. If this is the case, then the underlying science, which is fascinating in itself, may perhaps be being influenced by the debate around net zero.
Relating CO2 to the El Nino anomalies.
Data on sea surface temperatures can be downloaded from NOAA
Anomalous carbon dioxide concentrations in the time series from Mauna Loa can be obtained by fitting the model that includes the seasonal and yearly trend and extracting the residuals from it. These numbers represent the difference between the observed value and the expected value. The mean anomaly can then be calculated for each year. As the values for the sea surface temperature anomalies are comparable in size to those of the residuals from the CO2 concentration model the two time series can be plotted together.
A visual inspection does not suggest much of a relationship either in the values for each year or of any lagged effect. The direct correlation can be checked by fitting a regression to a scatter plot of the values for each year.
There is a very slight negative relationship, i.e in the opposite direction to the met office assumption. However this is not statistically significant.
Annual Yearly additions
The met office report adopted the approach of taking the mean carbon dioxide concentration for each year and looking at the change. In other words the yearly mean addition.
This measure shows a fairly consistent upward trend. However there is a very slight hint of periodicity in the residuals that could be worth investigating.
These changes can be plotted against the SST anomalies
Again the weak trend is not statistically significant and so of no value for prediction.
Conclusion
The claim by the Met office that predicted atmospheric CO2 levels are affected by the El Nino Southern Oscillation does not hold up well under independent analysis. If the effect was strong enough to be useful for prediction then it would have been easy to replicate using slightly different data sets derived from the same source (Scripps institute and NOAA). This is important, as if there were a strong el Nino effect it could have implications for conservation policy in the tropics. The reason that the Met office used the ENSO in their prediction of CO2 concentrations is unclear. There are many more direct effects that may alter anthropogenic emissions from land use change and fossil fuel burning. The Met office appear to be assuming that negative feedbacks occur if El Nino causes temporary warming *“Our prediction of a smaller atmospheric CO2 rise this year is due to natural climate variability being expected to temporarily strengthen natural carbon sinks, causing a temporary slowing of the rate of build-up of CO2 in the atmosphere.* Its unclear what this assumption is based on. If it were valid and could be shown to hold for longer term warming then it would act to prevent any”tipping point" due to anthropogenic warming. Unfortunately there does not seem to be strong empirical evidence to support it.
Epilogue
The global carbon cycle remains only partially understood. The increases in atmospheric CO2 concentrations are almost undoubtedly the result of emissions from burning fossil fuel. Land use change now plays a reduced role. However when emissions are measured and summed up, the numbers don’t seem to add up. Only around 50% of the emitted carbon dioxide remains in the atmosphere. Some is absorbed by the ocean, but an increasing proportion of this “missing sink” is presumed to be found in terrestrial ecosystems and attributable to storage in plants, soil and accumulating organic matter deposits such as peat bogs. Met office based their modelling on assumptions regarding the sensitivity of the sink to short term climate changes attributable to the ENSO. The downflux cannot be measured directly at a global scale so the sink strength is assumed to be the difference between the measured total emissions and the amount remaining in the atmosphere. If the sink strength was totally independent of emissions then net zero addition to the atmosphere could be attained by matching emissions to the sink. However it is not so simple. The terrestrial sink appears to have increased in strength at the same time as emissions have increased. One explanation is that carbon dioxide itself is resulting in greater rates of photosynthetic activity.
The data below is from (Friedlingstein et al. 2022) All values in billion tonnes of carbon per year (GtC/yr), for the globe.
The estimated strength of the terrestrial sink (Friedlingstein et al. 2022) is shown below.
The same data is shown as a figure on the global carbon project web site.
The current political emphasis on reducing carbon emissions at all cost makes it increasingly challenging to speculate on the underlying processes that produce these patterns in the available data. What does however seem clear is that the behavioural and economic changes that took place in 2020 in response to the pandemic had very little impact on global CO2 concentrations.
References
https://www.redalyc.org/pdf/339/33911302.pdf File = Met_office.Rmd